Allow int8 type in quantized_linear and quantized_fully_connected (#5900)

Summary:
Pull Request resolved: #5900

As titled.

Reviewed By: zonglinpeng

Differential Revision: D63659950

fbshipit-source-id: 1c28a797ef2a02ba86a2d860bf6ad44194f38ca0

Author: mcremon-meta

backends/cadence/aot/ops_registrations.py

@@ -111,7 +111,7 @@ def quantized_linear_meta(
     weight_size = list(weight.size())
     assert len(weight_size) == 2
     out_size[-1] = weight_size[0]
-    return src.new_empty(out_size, dtype=torch.uint8)
+    return src.new_empty(out_size, dtype=src.dtype)
 
 
 @register_fake("cadence::quantized_conv")